高阶异构数据层次联合聚类算法.pdfVIP

计算机研 究与发展 DOI：10．7544／issn1OO0～1239．2015 JournalofComputerResearchandDevelopment 52(1)：200—210，2015 高阶异构数据层次联合聚类算法 杨欣欣 黄少滨 (哈尔滨 ．[程大学计算机科学与技术学院 哈尔滨 150001) (yangxinxin051131@ 126．corn) A HierarchicalCo—ClusteringAlgorithm forHigh—OrderHeter0gene0usData YangXinxinandHuangShaobin (CollegeofComputerScienceandTechnology，HarbinEngineeringUniversity，Harbin150001) Abstract Theavailabilityofhigh—orderheterogeneousdatarepresentedwithmultiplefeaturescoming from heterogeneousdomainsisgettingmoreandmorecommoninrealworldapplication．H igh—order CO—clusteringalgorithmscan fusemultiple feature space information to improve clustering results effectivity。SOrecentlyitisbecomingoneofthehottestresearchtopics．M ostexistinghigh—orderco— clustering algorithms are non—hierarchical clustering algorithms． However， there are always hierarchicalclusterstructureshidden inhigh—orderheterogeneousdata．In ordertominethehidden patternsin datasetsmore effectively，we develop ahigh—order hierarchicalco—clustering algorithm (HHCC)．Goodman—Kruskalz-isusedtomeasuretheassociationofobjectsandfeatures。whichisan indexmeasuringassociationofcategoricalvariables．Theobjectswhicharestrongassociationare partitionedintothesameobjectsclusters，andsimutaneouslythefeatureswhicharestrongassociation arepartitioned into the same featuresclusterstoo． HHCC algorithm usesGoodman—Kruska1r to quantifythequalityofclusteringresultsofobjectsandfeaturesofeveryleve1．Accordingtooptimizing Goodman—Kruskalrbyalocallysearch approach，thenumberofclustersisautomaticallydetermined andclusteringresultsofeveryhierarchyareobtained．Thetop—down strategy isadoptedandatree— likeclusterstructure isformed at last． Experimentalresultsdemonstrate thatHHCC algorithm outperformsfour classicalhomogeneoushierarchicalalgorithmsand five previous high——order CO—— clustering